Closed mattwthompson closed 2 years ago
There's no point in bisecting, this fails a while back in the history. I arbitrarily went back to 0.10.1 and it fails. It also fails if I downgrade to 2020.2.4, 2020.2.0 and 2020.0.4 from there . Then, still on 2020.0.4, I dropped back to 0.10.0 and 0.9.2 and it fails with those version combinations as well.
The test itself is virtually untouched since 2019 or so. The only recent change I can find is here in #1036 but that doesn't seem aligned with the stereochemistry issue. The code in OpenEyeToolkitWrapper.to_openeye()
is functionally untouched for several months to a few years.
it fails I'm very confused as to how this was ever passing - I don't recall this test being turned off and on recently.
The problem here was that I'm not very smart:
atom_F = molecule.add_atom(35, 0, False, name="F")
Describe the bug Stereochemistry can be lost in
Molecule.to_openeye
in potentially weird cases. This test has been failing in CI for some amount of time.I have yet to bisect, so I don't know when this was introduced.To Reproduce
or this script, adapted from the unit test
Computing environment (please complete the following information): Based off of https://github.com/openforcefield/openff-toolkit/pull/1218 at 5fce37119ffe3720bf6177a1b483190bf41ec245
Additional context I don't think #1184 is relevant, not because I understand it deeply but because it has RDKit in the title and this behavior only errors out with
to_openeye
. Unless it should error out with both.I did only a cursory scan of OpenEye's release notes but nothing stood out to me.